We present a supervised machine learning framework for sequential datum-wise feature acquisition and classifier selection. The presented method sequentially acquires features during testing until it determines that ad...
详细信息
The indoor positioning for visually impaired people has influence on their daily life in unknown indoor environment. This study designs the robot that can assist the blind walking safety and navigate in indoor environ...
详细信息
To accommodate the wide range of input voltages supplied by redundant batteries and ensure an adequate hold-up time for communication systems during utility power failures, power supplies used in 5 G base stations typ...
详细信息
Microservices architecture is popular due to its scalability and flexibility. However, managing and troubleshooting distributed microservices-based systems can be challenging and time consuming. Auto-remediation of an...
详细信息
We study deep neural networks for the multi-label classification (M-lab) task through the lens of neural collapse (NC). Previous works have been restricted to the multi-class classification setting and discovered a pr...
详细信息
We study deep neural networks for the multi-label classification (M-lab) task through the lens of neural collapse (NC). Previous works have been restricted to the multi-class classification setting and discovered a prevalent NC phenomenon comprising of the following properties for the last-layer features: (i) the variability of features within every class collapses to zero, (ii) the set of feature means form an equi-angular tight frame (ETF), and (iii) the last layer classifiers collapse to the feature mean upon some scaling. We generalize the study to multi-label learning, and prove for the first time that a generalized NC phenomenon holds with the "pick-all-label" formulation, which we term as M-lab NC. While the ETF geometry remains consistent for features with a single label, multi-label scenarios introduce a unique combinatorial aspect we term the"tag-wise average" property, where the means of features with multiple labels are the scaled averages of means for single-label instances. Theoretically, under proper assumptions on the features, we establish that the only global optimizer of the pick-all-label cross-entropy loss satisfy the multi-label NC. In practice, we demonstrate that our findings can lead to better test performance with more efficient training techniques for M-lab learning. Copyright 2024 by the author(s)
We propose a new method called the N-particle underdamped Langevin algorithm for optimizing a special class of non-linear functionals defined over the space of probability measures. Examples of problems with this form...
We propose a new method called the N-particle underdamped Langevin algorithm for optimizing a special class of non-linear functionals defined over the space of probability measures. Examples of problems with this formulation include training mean-field neural networks, maximum mean discrepancy minimization and kernel Stein discrepancy minimization. Our algorithm is based on a novel spacetime discretization of the mean-field underdamped Langevin dynamics, for which we provide a new, fast mixing guarantee. In addition, we demonstrate that our algorithm converges globally in total variation distance, bridging the theoretical gap between the dynamics and its practical implementation. Copyright 2024 by the author(s)
The agriculture industry contributes most to expanding economies and populations and is essential to the production of high-quality food. Plant diseases are dependent on a variety of environmental variables that can s...
详细信息
In neural speech enhancement,a mismatch exists between the training objective,i.e.,Mean-Square Error(MSE),and perceptual quality evaluation metrics,i.e.,perceptual evaluation of speech quality and short-time objective...
详细信息
In neural speech enhancement,a mismatch exists between the training objective,i.e.,Mean-Square Error(MSE),and perceptual quality evaluation metrics,i.e.,perceptual evaluation of speech quality and short-time objective *** propose a novel reinforcement learning algorithm and network architecture,which incorporate a non-differentiable perceptual quality evaluation metric into the objective function using a dynamic filter *** the traditional dynamic filter implementation that directly generates a convolution kernel,we use a filter generation agent to predict the probability density function of a multivariate Gaussian distribution,from which we sample the convolution *** results show that the proposed reinforcement learning method clearly improves the perceptual quality over other supervised learning methods with the MSE objective function.
Language models are one of the biggest game changers in downstream NLP applications, especially in conversational agents. In spite of their awesome capabilities to generated responses to solve the inquiries, there are...
详细信息
The phenomenon of atmospheric haze arises due to the scattering of light by minute particles suspended in the atmosphere. This optical effect gives rise to visual degradation in images and videos. The degradation is p...
详细信息
The phenomenon of atmospheric haze arises due to the scattering of light by minute particles suspended in the atmosphere. This optical effect gives rise to visual degradation in images and videos. The degradation is primarily influenced by two key factors: atmospheric attenuation and scattered light. Scattered light causes an image to be veiled in a whitish veil, while attenuation diminishes the image inherent contrast. Efforts to enhance image and video quality necessitate the development of dehazing techniques capable of mitigating the adverse impact of haze. This scholarly endeavor presents a comprehensive survey of recent advancements in the domain of dehazing techniques, encompassing both conventional methodologies and those founded on machine learning principles. Traditional dehazing techniques leverage a haze model to deduce a dehazed rendition of an image or frame. In contrast, learning-based techniques employ sophisticated mechanisms such as Convolutional Neural Networks (CNNs) and different deep Generative Adversarial Networks (GANs) to create models that can discern dehazed representations by learning intricate parameters like transmission maps, atmospheric light conditions, or their combined effects. Furthermore, some learning-based approaches facilitate the direct generation of dehazed outputs from hazy inputs by assimilating the non-linear mapping between the two. This review study delves into a comprehensive examination of datasets utilized within learning-based dehazing methodologies, elucidating their characteristics and relevance. Furthermore, a systematic exposition of the merits and demerits inherent in distinct dehazing techniques is presented. The discourse culminates in the synthesis of the primary quandaries and challenges confronted by prevailing dehazing techniques. The assessment of dehazed image and frame quality is facilitated through the application of rigorous evaluation metrics, a discussion of which is incorporated. To provide empiri
暂无评论